[Core] Fix Ascend NPU discovery to support 8+ cards per node (#48543)

Signed-off-by: Xiaoshuang Liu <liuxiaoshuang4@huawei.com>
ray-project · Nov 5, 2024 · b2fafa3 · b2fafa3
1 parent a0dda94
commit b2fafa3
Show file tree

Hide file tree

Showing 2 changed files with 3 additions and 3 deletions.
diff --git a/python/ray/_private/accelerators/npu.py b/python/ray/_private/accelerators/npu.py
@@ -60,7 +60,7 @@ def get_current_node_num_accelerators() -> int:
             logger.debug("Could not import AscendCL: %s", e)
 
         try:
-            npu_files = glob.glob("/dev/davinci?")
+            npu_files = glob.glob("/dev/davinci[0-9]*")
             return len(npu_files)
         except Exception as e:
             logger.debug("Failed to detect number of NPUs: %s", e)

diff --git a/python/ray/tests/accelerators/test_npu.py b/python/ray/tests/accelerators/test_npu.py
@@ -11,8 +11,8 @@
 def test_autodetect_num_npus(mock_glob):
     with patch.dict(sys.modules):
         sys.modules["acl"] = None
-        mock_glob.return_value = [f"/dev/davinci{i}" for i in range(4)]
-        assert Accelerator.get_current_node_num_accelerators() == 4
+        mock_glob.return_value = [f"/dev/davinci{i}" for i in range(64)]
+        assert Accelerator.get_current_node_num_accelerators() == 64
 
 
 @patch("glob.glob")